Pre-training produces representations that are effective for a wide range of downstream tasks, but it is still unclear what properties of pre-training are necessary for effective gains. Notably, recent work shows that even pre-training on synthetic tasks can achieve significant gains in downstream tasks. In this work, we perform three experiments that iteratively simplify pre-training and show that the simplifications still retain much of its gains. First, building on prior work, we perform a systematic evaluation of three existing synthetic pre-training methods on six downstream tasks. We find the best synthetic pre-training method, LIME, attains an average of $67\%$ of the benefits of natural pre-training. Second, to our surprise, we find that pre-training on a simple and generic synthetic task defined by the set function achieves $65\%$ of the benefits, almost matching LIME. Third, we find that $39\%$ of the benefits can be attained by using merely the parameter statistics of synthetic pre-training.

Transfer learning and ensembling are two popular techniques for improving the performance and robustness of neural networks. Due to the high cost of pre-training, ensembles of models fine-tuned from a single pre-trained checkpoint are often used in practice. Such models end up in the same basin of the loss landscape, which we call the pre-train basin, and thus have limited diversity. In this work, we show that ensembles trained from a single pre-trained checkpoint may be improved by better exploring the pre-train basin, however, leaving the basin results in losing the benefits of transfer learning and in degradation of the ensemble quality. Based on the analysis of existing exploration methods, we propose a more effective modification of the Snapshot Ensembles (SSE) for transfer learning setup, StarSSE, which results in stronger ensembles and uniform model soups.

The Transformer and its variants have been proven to be efficient sequence learners in many different domains. Despite their staggering success, a critical issue has been the enormous number of parameters that must be trained (ranging from $10^7$ to $10^{11}$) along with the quadratic complexity of dot-product attention. In this work, we investigate the problem of approximating the two central components of the Transformer --- multi-head self-attention and point-wise feed-forward transformation, with reduced parameter space and computational complexity. We build upon recent developments in analyzing deep neural networks as numerical solvers of ordinary differential equations. Taking advantage of an analogy between Transformer stages and the evolution of a dynamical system of multiple interacting particles, we formulate a temporal evolution scheme, \name, to bypass costly dot-product attention over multiple stacked layers. We perform exhaustive experiments with \name\ on well-known encoder-decoder as well as encoder-only tasks. We observe that the degree of approximation (or inversely, the degree of parameter reduction) has different effects on the performance, depending on the task. While in the encoder-decoder regime, \name\ delivers performances comparable to the original Transformer, in encoder-only tasks it consistently outperforms Transformer along with several subsequent variants.

Deep neural networks have great representation power, but typically require large numbers of training examples. This motivates deep active learning methods that can significantly reduce the amount of labeled training data. Empirical successes of deep active learning have been recently reported in the literature, however, rigorous label complexity guarantees of deep active learning have remained elusive. This constitutes a significant gap between theory and practice. This paper tackles this gap by providing the first near-optimal label complexity guarantees for deep active learning. The key insight is to study deep active learning from the nonparametric classification perspective. Under standard low noise conditions, we show that active learning with neural networks can provably achieve the minimax label complexity, up to disagreement coefficient and other logarithmic terms. When equipped with an abstention option, we further develop an efficient deep active learning algorithm that achieves $\mathsf{polylog}(\frac{1}{\varepsilon})$ label complexity, without any low noise assumptions. We also provide extensions of our results beyond the commonly studied Sobolev/H\older spaces and develop label complexity guarantees for learning in Radon $\mathsf{BV}^2$ spaces, which have recently been proposed as natural function spaces associated with neural networks.

While ensembling deep neural networks has shown promise in improving generalization performance, scaling current ensemble methods for large models remains challenging. Given that recent progress in deep learning is largely driven by the scale, exemplified by the widespread adoption of large-scale neural network architectures, scalability emerges an increasingly critical issue for machine learning algorithms in the era of large-scale models. In this work, we first showcase the potential of low precision ensembling, where ensemble members are derived from a single model within low precision number systems in a training-free manner. Our empirical analysis demonstrates the effectiveness of our proposed low precision ensembling method compared to existing ensemble approaches.

Increasing the batch size is a popular way to speed up neural network training, but beyond some critical batch size, larger batch sizes yield diminishing returns. In this work, we study how the critical batch size changes based on properties of the optimization algorithm, including acceleration and preconditioning, through two different lenses: large scale experiments and analysis using a simple noisy quadratic model (NQM). We experimentally demonstrate that optimization algorithms that employ preconditioning, specifically Adam and K-FAC, result in much larger critical batch sizes than stochastic gradient descent with momentum. We also demonstrate that the NQM captures many of the essential features of real neural network training, despite being drastically simpler to work with. The NQM predicts our results with preconditioned optimizers, previous results with accelerated gradient descent, and other results around optimal learning rates and large batch training, making it a useful tool to generate testable predictions about neural network optimization.

Pre-training produces representations that are effective for a wide range of downstream tasks, but it is still unclear what properties of pre-training are necessary for effective gains. Notably, recent work shows that even pre-training on synthetic tasks can achieve significant gains in downstream tasks. In this work, we perform three experiments that iteratively simplify pre-training and show that the simplifications still retain much of its gains. First, building on prior work, we perform a systematic evaluation of three existing synthetic pre-training methods on six downstream tasks. We find the best synthetic pre-training method, LIME, attains an average of 67\% of the benefits of natural pre-training.

We derive asymptotically precise expressions for test and train errors of denoising score matching (DSM) in generative diffusion models. The score function is parameterized by random features neural networks, with the target distribution being $d$-dimensional standard Gaussian. We operate in a regime where the dimension $d$, number of data samples $n$, and number of features $p$ tend to infinity while keeping the ratios $\psi_n=\frac{n}{d}$ and $\psi_p=\frac{p}{d}$ fixed. By characterizing the test and train errors, we identify regimes of generalization and memorization in diffusion models. Furthermore, our work sheds light on the conditions enhancing either generalization or memorization. Consistent with prior empirical observations, our findings indicate that the model complexity ($p$) and the number of noise samples per data sample ($m$) used during DSM significantly influence generalization and memorization behaviors.

EDIT: The authors response addresses my concerns about reproducibility and the appendix they promise should contain the missing details of their setup. Summary and relation to previous work: This paper presents projection weighted canonical correlation analysis (CCA) as a method to interrogate neural network representations. It is a direct continuation of previous work on singular vector canonical correlation analysis (SVCCA), addressing several limitations of SVCCA and applying the method to new models and questions. They observe that some CCA components stabilize early in training and remain stable (signal) while other components continue to vary even after performance has converged (noise). This motivates replacing the mean canonical correlation coefficient (used as a similarity metric in SVCCA) with a weighted mean, where components that better explain the activation patterns receive a higher weight.